Ryan Tandy [Fri, 30 Oct 2020 01:36:26 +0000 (01:36 +0000)]
Do not call gnutls_global_set_mutex()
Bug-Debian: https://bugs.debian.org/803197
Forwarded: no
Since GnuTLS moved to implicit initialization on library load, calling
this function deinitializes GnuTLS and then re-initializes it.
When GnuTLS uses /dev/urandom as an entropy source (getrandom() not
available, or older versions of GnuTLS), and the application closed all
file descriptors at startup, this could result in GnuTLS opening
/dev/urandom over one of the application's file descriptors when
re-initialized.
Additionally, the custom mutex functions are never reset, so if libldap
is unloaded (for example via dlclose()) after calling this, its code
may be unmapped and the application could crash when GnuTLS calls the
mutex functions.
The default behaviour of GnuTLS, using pthreads, should be suitable on
all Debian systems, and is probably the same as what libldap uses
anyway.
OpenLDAP upstream conservatively assumes that any change to the version
number of libdb can result in an API-breaking change that could impact
the database. In Debian, we know that such changes require bumping the
library soname and changing the package name, and demand such rigor from
our package maintainers even when upstreams don't deliver; so any such
check in the source code works against the packaging system by forcing
database upgrades when we know none are required. Disable this check
so we rely on the packaging system to do its job.
Switch to lt_dlopenadvise() so back_perl can be opened with RTLD_GLOBAL. Open all modules with RTLD_GLOBAL, needed so that back_perl can load non-trivial Perl extensions that require symbols from back_perl.so itself.
Bug-Debian: http://bugs.debian.org/327585
Gbp-Pq: Name switch-to-lt_dlopenadvise-to-get-RTLD_GLOBAL-set.diff
Rip out code that second-guesses the libsasl soname / Debian shlibs. If
cyrus sasl upstream is breaking the ABI, this needs to be fixed upstream
there, not kludged around upstream here!
Steve Langasek [Fri, 30 Oct 2020 01:36:26 +0000 (01:36 +0000)]
getaddrinfo-is-threadsafe
OpenLDAP upstream conservatively assumes that certain resolver functions
(getaddrinfo, getnameinfo, res_query, dn_expand) are not re-entrant; but we
know that the glibc implementations of these functions are thread-safe, so
we should bypass the use of this mutex. This fixes a locking problem when
an application uses libldap and libnss-ldap is also used for hosts
resolution.
Closes Debian bug #340601.
Not suitable for forwarding upstream; might be made suitable by adding a
configure-time check for glibc and disabling the mutex only on known
thread-safe implementations.
Add symbol versioning to the public LDAP libraries. This is required for
library transitions, such as the current transition from 2.1 to 2.4,
since programs will sometimes have both libraries loaded by different
dependency chains during the transition.
Not yet contributed upstream.
Upstream ITS #5365 filed requesting symbol versioning for libldap and
libber.
Document in the man page that slapindex should be run as the same user
as slapd, and print a warning if it's run as root (since Debian defaults
to running slapd as openldap).
Not suitable for upstream in this form. This patch needs to be reworked
to check the BerkeleyDB database ownership and only warn if running as
root with a database that's not owned by root.
Upstream ITS #5356 filed requesting better handling of this. Current
upstream discussion leans towards putting the check into the database
backend and aborting if slapd is run as a different user than the database
owner, which is an even better fix.
Patch from evolution-exchange (2.10.3). The ldap_ntlm_bind function is
actually called by evolution-data-server, checked at version 1.12.2.
Without this patch, the Exchange addressbook integration uses simple binds
with cleartext passwords.
Russ checked with openldap-software for upstream's opinion on this patch
on 2007-12-21. Upstream had never received it as a patch submission and
given that it's apparently only for older Exchange servers that can't do
SASL and DIGEST-MD5, it's not very appealing.
Bug#457374 filed against evolution-data-server asking if this support is
still required on 2007-12-21.
Ryan Tandy [Mon, 20 Apr 2020 18:19:54 +0000 (19:19 +0100)]
Do not call gnutls_global_set_mutex()
Bug-Debian: https://bugs.debian.org/803197
Forwarded: no
Since GnuTLS moved to implicit initialization on library load, calling
this function deinitializes GnuTLS and then re-initializes it.
When GnuTLS uses /dev/urandom as an entropy source (getrandom() not
available, or older versions of GnuTLS), and the application closed all
file descriptors at startup, this could result in GnuTLS opening
/dev/urandom over one of the application's file descriptors when
re-initialized.
Additionally, the custom mutex functions are never reset, so if libldap
is unloaded (for example via dlclose()) after calling this, its code
may be unmapped and the application could crash when GnuTLS calls the
mutex functions.
The default behaviour of GnuTLS, using pthreads, should be suitable on
all Debian systems, and is probably the same as what libldap uses
anyway.
OpenLDAP upstream conservatively assumes that any change to the version
number of libdb can result in an API-breaking change that could impact
the database. In Debian, we know that such changes require bumping the
library soname and changing the package name, and demand such rigor from
our package maintainers even when upstreams don't deliver; so any such
check in the source code works against the packaging system by forcing
database upgrades when we know none are required. Disable this check
so we rely on the packaging system to do its job.
Switch to lt_dlopenadvise() so back_perl can be opened with RTLD_GLOBAL. Open all modules with RTLD_GLOBAL, needed so that back_perl can load non-trivial Perl extensions that require symbols from back_perl.so itself.
Bug-Debian: http://bugs.debian.org/327585
Gbp-Pq: Name switch-to-lt_dlopenadvise-to-get-RTLD_GLOBAL-set.diff
Rip out code that second-guesses the libsasl soname / Debian shlibs. If
cyrus sasl upstream is breaking the ABI, this needs to be fixed upstream
there, not kludged around upstream here!
Steve Langasek [Mon, 20 Apr 2020 18:19:54 +0000 (19:19 +0100)]
getaddrinfo-is-threadsafe
OpenLDAP upstream conservatively assumes that certain resolver functions
(getaddrinfo, getnameinfo, res_query, dn_expand) are not re-entrant; but we
know that the glibc implementations of these functions are thread-safe, so
we should bypass the use of this mutex. This fixes a locking problem when
an application uses libldap and libnss-ldap is also used for hosts
resolution.
Closes Debian bug #340601.
Not suitable for forwarding upstream; might be made suitable by adding a
configure-time check for glibc and disabling the mutex only on known
thread-safe implementations.
Add symbol versioning to the public LDAP libraries. This is required for
library transitions, such as the current transition from 2.1 to 2.4,
since programs will sometimes have both libraries loaded by different
dependency chains during the transition.
Not yet contributed upstream.
Upstream ITS #5365 filed requesting symbol versioning for libldap and
libber.
Document in the man page that slapindex should be run as the same user
as slapd, and print a warning if it's run as root (since Debian defaults
to running slapd as openldap).
Not suitable for upstream in this form. This patch needs to be reworked
to check the BerkeleyDB database ownership and only warn if running as
root with a database that's not owned by root.
Upstream ITS #5356 filed requesting better handling of this. Current
upstream discussion leans towards putting the check into the database
backend and aborting if slapd is run as a different user than the database
owner, which is an even better fix.
Patch from evolution-exchange (2.10.3). The ldap_ntlm_bind function is
actually called by evolution-data-server, checked at version 1.12.2.
Without this patch, the Exchange addressbook integration uses simple binds
with cleartext passwords.
Russ checked with openldap-software for upstream's opinion on this patch
on 2007-12-21. Upstream had never received it as a patch submission and
given that it's apparently only for older Exchange servers that can't do
SASL and DIGEST-MD5, it's not very appealing.
Bug#457374 filed against evolution-data-server asking if this support is
still required on 2007-12-21.
Switch to lt_dlopenadvise() so back_perl can be opened with RTLD_GLOBAL. Open all modules with RTLD_GLOBAL, needed so that back_perl can load non-trivial Perl extensions that require symbols from back_perl.so itself.
Gbp-Pq: Name switch-to-lt_dlopenadvise-to-get-RTLD_GLOBAL-set.diff
Rip out code that second-guesses the libsasl soname / Debian shlibs. If
cyrus sasl upstream is breaking the ABI, this needs to be fixed upstream
there, not kludged around upstream here!
OpenLDAP upstream conservatively assumes that certain resolver functions
(getaddrinfo, getnameinfo, res_query, dn_expand) are not re-entrant; but we
know that the glibc implementations of these functions are thread-safe, so
we should bypass the use of this mutex. This fixes a locking problem when
an application uses libldap and libnss-ldap is also used for hosts
resolution.
Closes Debian bug #340601.
Not suitable for forwarding upstream; might be made suitable by adding a
configure-time check for glibc and disabling the mutex only on known
thread-safe implementations.
Add symbol versioning to the public LDAP libraries. This is required for
library transitions, such as the current transition from 2.1 to 2.4,
since programs will sometimes have both libraries loaded by different
dependency chains during the transition.
Not yet contributed upstream.
Upstream ITS #5365 filed requesting symbol versioning for libldap and
libber.
Document in the man page that slapindex should be run as the same user
as slapd, and print a warning if it's run as root (since Debian defaults
to running slapd as openldap).
Not suitable for upstream in this form. This patch needs to be reworked
to check the BerkeleyDB database ownership and only warn if running as
root with a database that's not owned by root.
Upstream ITS #5356 filed requesting better handling of this. Current
upstream discussion leans towards putting the check into the database
backend and aborting if slapd is run as a different user than the database
owner, which is an even better fix.
Patch from evolution-exchange (2.10.3). The ldap_ntlm_bind function is
actually called by evolution-data-server, checked at version 1.12.2.
Without this patch, the Exchange addressbook integration uses simple binds
with cleartext passwords.
Russ checked with openldap-software for upstream's opinion on this patch
on 2007-12-21. Upstream had never received it as a patch submission and
given that it's apparently only for older Exchange servers that can't do
SASL and DIGEST-MD5, it's not very appealing.
Bug#457374 filed against evolution-data-server asking if this support is
still required on 2007-12-21.
* Fix slapd to restrict rootDN proxyauthz to its own databases
(CVE-2019-13057) (ITS#9038) (Closes: #932997)
* Fix slapd to enforce sasl_ssf ACL statement on every connection
(CVE-2019-13565) (ITS#9052) (Closes: #932998)
* Fix slapo-rwm to not free original filter when rewritten filter is invalid
(ITS#8964) (Closes: #934277, LP: #1838370)